GitHub - LewisLiuPub/DeepICP

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
configs		configs
pointconv		pointconv
tf_ops		tf_ops
.gitignore		.gitignore
ICP.ipynb		ICP.ipynb
LossFunctionTest.ipynb		LossFunctionTest.ipynb
RandomStuff.ipynb		RandomStuff.ipynb
Untitled.ipynb		Untitled.ipynb
args_options.py		args_options.py
config.py		config.py
dataset_iterator.py		dataset_iterator.py
history_callback.py		history_callback.py
icp.py		icp.py
index_dataset.py		index_dataset.py
loss.py		loss.py
model.py		model.py
net_builder.py		net_builder.py
readme.txt		readme.txt
realsense_utils.py		realsense_utils.py
test_iterator.py		test_iterator.py
tf_utils.py		tf_utils.py
train.py		train.py

Repository files navigation

python 3.6
tensorflow >= 2.2
CUDA 10.1

--------------

DO NOT USE CONDA as linking tensorflow for compiling PointNet++ doesn't work out of box

--------------

create 
./configs/project_config.json
using
./configs/project_config_template.json
and set up your paths config file

-------------

DATASET
https://apollo.auto/southbay.html?fbclid=IwAR3qENSb6C3dDpjjeVM-Fn9kM_m7vaidhIFGzj7tzqIsQp3aOpKjQaRRfoA
download at the bottom of page, place pcds and poses in folder

--------------

depending on the way you have your cuda setup you might need to run this in terminal

export LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
https://stackoverflow.com/a/64472380 for more troubleshooting

--------------

to compile PointNet++ ops:
sh tf_ops/compile_ops.sh

--------------

python3 index_datset.py

--------------

to create new default model

python3 train.py -mn <model_name> -new

--------------

for for more options

python3 train.py --help

create new configs in their respecive folders to use them as arguments